Halifax November: -0.9% MoM -0.6% QoQ -1.0% YoY

System

Let's sum it up with a picture.



Of course, they're both probably !!!!!, but I'm enjoying playing devil's advocate and exploiting the awkward position geneer has put all his buddies in.

Edit: Aww there's a typo in my graph

joguest

Joeskeppi wrote: »

Let's sum it up with a picture.



Of course, they're both probably !!!!!, but I'm enjoying playing devil's advocate and exploiting the awkward position geneer has put all his buddies in.

Edit: Aww there's a typo in my graph

OK, so which part of me saying the Nationwide being unusually less accurate (in tracking the LR) at the moment did you not understand? You have just proved that the Nationwide is usually more accurate in tracking the LR (well, since early 2009), which is entirely consistent with everything I've said. However, over the last twelve months, the Halifax better tracks the LR trend.

You don't seem to realise that the graph is in itself normalised to peak prices (to provide an interesting comparison). If it's normalised to prices a year ago then it would look quite different.

wotsthat

joguest wrote: »

Does that mean taking a single Nationwide YoY, a single Halifax YoY and deciding that Halifax is more accurate because it's closer to a single LR YoY is good stats then?

I didn't say it makes the Halifax more accurate. That is an absolute statement. This is what I did say:

"The Nationwide is around 5% higher YoY than the LR (-3.2% YoY), so one would assume the 'fax is currently more reliable. "

"At the moment, the nationwide is (unusually) the most out of step with the LR figures (the only figures that really count). "

Notice the words:
'assume' - an assumption is not the same thing as claiming to have proved a hypothesis
'currently' - meaning at this point in time - i.e. it's not a claim that the Halifax is more accurate throughout its history
'unusually' - this implies that I think the Nationwide is usually more accurate - more accurate in the sense that it better tracks the LR data

Yes, it is good stats to compare the YoY figures between the three and come to the conclusion that, over that time period (which is more meaningful for purposes of looking at purchasing /selling decisions for houses), the Halifax is showing a better agreement with the LR. If I'd made some wild, speculative hypothesis on the basis of the observation then you would be right to tear me apart. However, I didn't - all I did was state (not unreasonably) that the Halifax YoY is showing a closer agreement to the LR YoY (the figures based on actual sales, that we assume to be most accurate) than the Nationwide, which has drifted uncharacteristically to a ~5% difference in the trend over the last 12 months (albeit the LR is delayed).

I suspect the gap will close over the next few months.

Would it be fair to say the Nationwide figure from a year ago was lower than LR so noise (sorry standard error) means that it's pretty pointless trying to base an argument that Halifax is more reliable based on 3 YoY figures.

Apart from some noise ('uncharateristic difference in the trend) LR & Nationwide look closely correlated.

When they stay correlated I'll be saying just that rather than calling it a reversion to trend if that's ok.

Cleaver

joguest wrote: »

the normalised standard error for the percentage change in house prices between two points in time decreases with the inverse of the time period.

I have a faulty flux-capacitor at the moment and you may very well be the man to take a look at it.

joguest

Cleaver wrote: »

I have a faulty flux-capacitor at the moment and you may very well be the man to take a look at it.

Excellent - send it my way.

The normalised standard error (there is no noise that I'm aware of - noise would be an unwanted, random, external addition to the data) should decrease with increasing time frame, provided that house prices changes are larger over the long term than the short term (which isn't actually true at the moment).

Heyman_2

Cleaver wrote: »

I have a faulty flux-capacitor at the moment and you may very well be the man to take a look at it.

I usually find a good whack on the dashboard sorts mine out. I went back in time to 1990 yesterday and bought myself a nice cheap house, happy days :beer:

joguest

wotsthat wrote: »

Would it be fair to say the Nationwide figure from a year ago was lower than LR so noise (sorry standard error) means that it's pretty pointless trying to base an argument that Halifax is more reliable based on 3 YoY figures.

Apart from some noise ('uncharateristic difference in the trend) LR & Nationwide look closely correlated.

When they stay correlated I'll be saying just that rather than calling it a reversion to trend if that's ok.

How do you know the 'uncharacteristic difference in the trend' is 'noise', as you put it?

The Halifax trend was well correlated with both the LR and Nationwide trends prior to 2009. Why shouldn't it be well correlated again? Has the Halifax been uncharacteristic since 2009?

The data is what the data is. In the same way the Nationwide tracked the LR better since 2009, we can also say that the Halifax has tracked the LR better over the last 12 months. That may well change in the coming months, but if you're arguing that the Nationwide is more accurate over the last 2.5 years because it tracks the LR more closely, then the same argument applies to saying the Halifax is more accurate in tracking the LR over the last 12 months (at the moment). You can't make one argument and ignore the other just because it doesn't suit your prejudice.

wotsthat

joguest wrote: »

The data is what the data is. In the same way the Nationwide tracked the LR better since 2009, we can also say that the Halifax has tracked the LR better over the last 12 months. That may well change in the coming months, but if you're arguing that the Nationwide is more accurate over the last 2.5 years because it tracks the LR more closely, then the same argument applies to saying the Halifax is more accurate in tracking the LR over the last 12 months (at the moment). You can't make one argument and ignore the other just because it doesn't suit your prejudice.

My argument is consistent. You can't take three YoY figures from different indexes and make any credible conclusion with them either way.

If you want to look at the tracking though...

I wouldn't try and argue that Halifax is tracking LR better over the last 12 months because as you say the normalised standard error decreases with increasing time frame and I wonder if 12 months is along enough time period to allow such a conclusion.

Nationwide though has tracked LR closely for the whole of the six years of your graph. Would that be long enough to reduce the standard error to acceptable levels?

malkie76

Just to point out a mathematical thing; it's impossible to compare without knowing what the 100% value was for each Registry, or an understanding on how accurate or representative the value was.
Furthermore, just by looking at the graph it's clear that different months were used to give the 100% value (as each registry would have peaked at a different month) making any comparison rather flawed.

joguest

wotsthat wrote: »

My argument is consistent. You can't take three YoY figures from different indexes and make any credible conclusion with them either way.

If you want to look at the tracking though...

I wouldn't try and argue that Halifax is tracking LR better over the last 12 months because as you say the normalised standard error decreases with increasing time frame and I wonder if 12 months is along enough time period to allow such a conclusion.

Nationwide though has tracked LR closely for the whole of the six years of your graph. Would that be long enough to reduce the standard error to acceptable levels?

It all depends on where you to decide to normalise the data to. If we zoom out on the previous plot and show the data going back to 2001 then it becomes apparent that the LR doesn't track the other two that closely before ~2006.



However, if we now plot the same data, but normalised to the mean price over the data-set, then the LR does track the other two more closely before ~2006. Additionally (if this normalisation is to be better believed), then the Halifax seems to overestimate the 2007 peak price, which accounts for how the Nationwide appears to track the LR more closely since 2009 in the previous plot. Indeed, in the plot below, the LR is between the two other plots since 2009, suggesting that the Nationwide is slightly overestimating prices and the Halifax is slightly underestimating prices. However, at the moment, the Halifax is comparatively closer to the LR, which backs up the idea that the Halifax is more accurately reflecting changes in house prices over the last year.